Modeling and Visualization for Polymers, Surfaces and Biomolecules
Abstract
The primary aim of the project was to develop practical models and algorithms for robust optimization. Towards this goal, we have improved the efficiency of linear and nonlinear algorithms for solving robust optimization models. By specializing the ordering of the key matrix (ADAt), we have reduced the computational times for factorizations -- by over 100 times for larger examples. The largest LP problem solved to date (with 16,000 scenarios) consists of approximately 1 million constraints and 1.7 million variables. More importantly, the run time is a linear function of the number of scenarios. Hence the primary bottleneck for solving large examples is the amount of available computer memory. This result applies to a spectrum of planning problems since the ordering routine does not take advantage of the matrix structure within a scenario. Over the past several years, we have continued to specialize the large-scale optimization algorithms. Also, we have worked on the selection of the scenarios for robust optimization so that the number of scenarios is kept to a reasonable level. The use of out-of-sample precision tests have been designed and tested for evaluating the confidence in the recommendations of the models.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 15, 1997
- Accession Number
- ADA334683
Entities
People
- John M. Mulvey
- Robert J. Vanderbei
Organizations
- Princeton University